Re: Request to register "identifier" relation type

[Herbert Van de Sompel <[e-mail]>]

On Aug 9, 2017, at 18:59, Peter Williams <[e-mail]> wrote:
> 
> This discussion has convinced me that no existing relation is a good match for the proposed semantics. However, i am concerned that the proposed relation has taken this much discussion to understand. The confusion it generated here does not bode well for it being used properly by mere mortals. 
> 

Personally, I think the confusion was largely caused by thinking from the perspective that this was canonical or bookmark, and us having to show it was not.

I hope that a reading of the I-D itself, without coming from the canonical/bookmark perspective does make it clear what "identifier" is about. Various scenarios illustrate what it is intended for, and, IMO, the short description "preferred for referencing" is clear too. 

> I think a more concrete name would greatly improve the usability. Finding a more concrete name will be challenging because this relation conflate several different relationships into a single name. These different semantics are called out in the I-D in sections 3.1-3.4.
> 

The semantics are the same in all scenarios: the target URI is preferred for referencing. 

> Having multiple relations, one for each semantic, might be another way to address the usability issues. Some of those use cases seem to be covered by existing relations. For example, `canonical` seems tailor made for the "Version Identifiers" use case.

Our blog post shows that "canonical" is not appropriate at all for the Wikipedia versioning case. They want the generic URI (current version) to be indexed - canonical. They want the version-specific URI to be referenced - identifier. 

Greetings

Herbert

> For other use cases, such as "Multi-Resource Publications" and "Persistent Identifiers", there don't seem to be any existing relations that would work. Relations for those narrower use cases would be much easier to understand and use.
> 
> Peter
> 
>> On Wed, Aug 9, 2017 at 10:17 AM, Ed Summers <[e-mail]> wrote:
>> Hi Herbert,
>> 
>> > On Aug 9, 2017, at 10:35 AM, Herbert Van de Sompel <[e-mail]> wrote:
>> >
>> > * On August 5, Ed Summers posted a question regarding applying "bookmark" to <link> to the WHATWG list, see https://lists.w3.org/Archives/Public/public-whatwg-archive/2017Aug/0001.html. There are no responses to this post, so far.
>> 
>> There have been a few responses if you look at the list of emails for August:
>> 
>>     https://lists.w3.org/Archives/Public/public-whatwg-archive/2017Aug/
>> 
>> > * On August 9, Ed Summer posted a similar question to WHATWG/HTML GitHub, see https://github.com/whatwg/html/issues/2899. There is a reaction from @annevk who (1) speculates that the reason "bookmark" is not to be used with <link> might be in order not to overlap with "canonical" (2) suggests the use of "canonical" :-)
>> 
>> Yes, canonical seems to be the relation that most people are reaching for initially. I did myself on reading your I-D. The fact that seasoned hands like Kevin Marks and Anne van Kesteren are as well says something.
>> 
>> > * Michael Nelson has further explored "bookmark" and has confirmed that there effectively is a reason for not allowing "bookmark" in <link>. It is related to its target use case: surfacing a link for content contained in a *part* of a page. Hence, Michael concludes that making "bookmark" usable with <link> will most likely not happen. @annevk's GitHub response does not seem to contradict that. Michael based his findings on studying http://tantek.com/log/2002/11.html#L20021128t1352 and https://html.spec.whatwg.org/multipage/links.html#link-type-bookmark. He may write another blog post about this, but, for now, here's how he explained on Twitter https://twitter.com/i/moments/895081563653902336
>> 
>> Yes, it looks like that's probably where things will sit. As Anne indicated it's likely that rel=bookmark cannot be used with <link> because of perceived confusion it would cause with canonical. The semantics of parts of pages vs the page itself don't seem terribly significant to me from an implementation perspective. Unfortunately 'identifier' will also probably cause some confusion as well. As systems that rely on 'identifier' get developed that will be something for them to deal with.
>> 
>> Thanks for considering all the questions and tracking the conversation over on the WHATWG list. It speaks to the spirit of what you all are trying to achieve with this I-D.
>> 
>> //Ed
>